<result> <column_name>rem_sleep_pct</column_name> <family>uncertain</family> <subtype>none</subtype> <confidence>0.45</confidence> <requires_user_confirmation>true</requires_user_confirmation> <sum_meaningful>false</sum_meaningful> <signed>false</signed> <reducers> <primary_reducer>mean</primary_reducer> <valid_reducers>mean, median, min, max, distribution</valid_reducers> <excluded_reducers>sum (a per-entity percentage share does not accumulate), product (values are about 19, not a compounding factor near 1.0)</excluded_reducers> <average_semantics>arithmetic_mean</average_semantics> </reducers> <inner_outer_measure> <inner_measure_reducer>min</inner_measure_reducer> <outer_measure_reducer>max</outer_measure_reducer> </inner_outer_measure> <multiplicative_partner_derivable>false</multiplicative_partner_derivable> <reasoning> The table is a DIMENSION with no sequence axis, so multiplicative is highly implausible and no cumsum or cumprod reconstruction is possible. The "pct" token is the only multiplicative signal. The data contradicts it: the mean and median are about 19.1, the range is 9.6 to 27.0, nothing is near 1.0, and the log-symmetry gain is negative. The values look like a bounded percentage-of-sleep level, which in practice behaves as an intensive level where mean is the natural rollup. Because the name is the only multiplicative evidence and it conflicts with the numbers, the family is uncertain, and the reducers shown are the safe intensive-style set pending confirmation. </reasoning> <table_type_conflict>false</table_type_conflict> </result>